Data augmentation in PyTorch - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is data augmentation in machine learning?

Data augmentation is a technique to increase the size and diversity of training data by making small changes to existing data, like flipping or rotating images. This helps models learn better and avoid overfitting.

beginner

Name two common image data augmentation techniques.

Two common image data augmentation techniques are flipping (horizontal or vertical) and rotation by small angles. These create new images that help the model see different views of the same object.

intermediate

How does data augmentation help prevent overfitting?

Data augmentation adds variety to training data, so the model doesn't memorize exact examples. This makes the model generalize better to new data, reducing overfitting.

beginner

Show a simple PyTorch code snippet to apply random horizontal flip to images during training.

from torchvision import transforms
transform = transforms.Compose([
    transforms.RandomHorizontalFlip(p=0.5),
    transforms.ToTensor()
])

This code flips images horizontally with 50% chance before converting them to tensors.

intermediate

What is the difference between online and offline data augmentation?

Offline augmentation creates new data files before training, increasing dataset size on disk. Online augmentation applies random changes on the fly during training, saving storage and adding variety each epoch.

Which of the following is NOT a typical data augmentation technique for images?

A Random rotation

B Horizontal flip

C Changing image file format

D Adding Gaussian noise

Changing the image file format does not augment the image content or variety; it only changes how the image is stored.

Why do we use data augmentation in training machine learning models?

A To reduce training time

B To increase dataset size and variety

C To make the model smaller

D To remove noisy data

Data augmentation increases the size and diversity of training data, helping models learn better.

In PyTorch, which module provides common data augmentation transforms?

A torch.utils.data

B torch.optim

C torch.nn

D torchvision.transforms

torchvision.transforms contains many useful image augmentation functions.

What does RandomHorizontalFlip(p=0.5) do during training?

A Flips images horizontally with 50% chance

B Flips every image horizontally

C Flips images vertically with 50% chance

D Does nothing

It flips each image horizontally with a probability of 0.5.

Which is a benefit of online data augmentation over offline augmentation?

A Adds variety each training epoch

B Requires more disk space

C Slows down training significantly

D Creates fixed augmented dataset

Online augmentation applies random changes during training, so data varies each epoch.

Explain what data augmentation is and why it is useful in training machine learning models.

Think about how changing images slightly can help a model learn better.

Describe how you would implement data augmentation in a PyTorch image classification project.

Focus on the code steps to add augmentation before feeding images to the model.